We rebuilt a major Indian bank's tech platform to handle 100x more loan transactions.
Discover how Pinsaar transformed one of the largest private banks' loan platform in just 12 months
1. Bottleneck Identified
The bank's legacy loan engine—an aging monolith stitched together over several upgrade cycles—was designed for branch-centric traffic in the early 2000s. In 2024 it could push only ~5 completed transactions per second across four loan categories and one EMI product. Every payday, festival campaign, or interest-rate promo triggered five-minute waits, manual queue juggling at branches, and end-of-day batches that routinely overflowed into the next morning. IT ops filed more than 3,000 Sev-2 tickets in the prior fiscal year just to keep the lights on.
When a prospective borrower taps "Apply" and the spinner hangs, they feel ignored—and bounce to a fintech with one-click approval. The bottleneck harmed brand perception, dragged Net Promoter Score down by 11 points, and left tens of crores in fee income on the table each quarter. Front-line staff drowned in exception handling instead of cross-selling; auditors flagged the growing backlog as an operational risk; and the CTO's roadmap for new digital products was effectively frozen.
2. Vision & Green-Light
Pinsaar engaged over 4 months in discovery workshops, white-board sessions, and live demos with eight distinct stakeholder groups—IT Infrastructure, Business Transformation, Audit, InfoSec, Core Banking, Risk, DBA, and Digital Channels. Together we translated the pain into a single, hard goal: sustain 500 transactions per second, bank-wide, inside 12 months, with no regression in compliance or security posture. The steering committee signed off after witnessing a proof-of-concept that hit 200 TPS on commodity hardware.
A unified, numeric KPI cut through org politics: if a proposed feature didn't move the needle toward 500 TPS or compliance, it was deferred. The bank could now align budgets, vendor contracts, and internal OKRs around one North Star, ensuring momentum and eliminating scope creep that had sunk previous modernisation attempts.
3. Architecture Sprint
Chief Enterprise Architect Anuraag Vaidya led a two-week intensive to break the goal into four non-negotiables: 1. Speed – ≥ 500 TPS at p99 < 250 ms. 2. Always-on – active-active across dual Tier-IV data centres. 3. Open-source core – enterprise-licensed but transparent and audit-ready. 4. Click-once deployment – full rollback in < 5 minutes. We mapped 14 existing system touch-points, catalogued 127 third-party dependencies, and produced an executable ADR (Architecture Decision Record) library to keep future debates grounded.
Writing requirements as measurable SLOs meant every engineer, tester, and vendor could self-validate. It also armed Procurement with concrete acceptance criteria, accelerating contract cycles by 30 %.
4. Blueprint Signed-Off
The final design introduced a two-layer model: • Orchestrator tier—stateless HTTPS gateways terminating public traffic, handling auth, throttling, and format transforms. • Micro-service tier—gRPC workers focused on single-purpose domains (offering, underwriting, disbursal, ledger). Data strategy blended Oracle NoSQL (flexible schema, strong consistency) with a Redis Cluster for sub-millisecond master-data reads. All logs and metrics streamed into segregated ELK pipelines: business-facing and ops-facing. The platform spanned two mirrored data centres, each capable of handling 70 % of peak on its own.
Layering let us dial up capacity by adding orchestrators without touching core logic, while gRPC trimmed network overhead by ~60 % compared to REST. Active-active topology ensured a DC-wide outage became a non-event for end users, satisfying the board's new Business Continuity mandate.
5. Automation Edge
Pinsaar leveraged its in-house Data Automation Language (DAL)—a DSL where engineers declare entities, caching rules, privacy tags, and validation flows in YAML-like syntax. DAL compiled these declarations into: • Fully-typed gRPC service stubs • Orchestrator routes with RBAC guards • Cache-aside wrappers for Redis • Structured log emitters • Terraform snippets for infrastructure This slashed boilerplate by 60 %, enabling a lean team of 18 engineers to ship the equivalent of 45 FTEs of output.
Generated code is consistent code: fewer edge-case bugs, predictable performance, and rapid onboarding of new developers. It also meant that late regulatory tweaks—like updated KYC fields—were a matter of editing one DAL file and redeploying in under an hour.
6. Build & Integrate
Implementation spanned 64 servers (Kubernetes on RHEL) with CI/CD pipelines powered by GitLab-CI and Ansible. Over 100 enterprise APIs (KYC, AML, credit bureau, e-sign, RTGS/IMPS) were encapsulated into idempotent connectors. 50 watchdog micro-jobs monitored queues, retried failed calls, and raised Slack-Opsgenie alerts with contextual runbook links. Code quality gates (SonarCloud) blocked any merge below 90 % coverage or with critical CVEs.
Automation shifted defect detection left: 83 % of issues surfaced in CI, saving an estimated ₹2.1 crore in rework and unplanned overtime. The bank's audit team gained real-time visibility into pipeline artifacts, satisfying SOX-style traceability without extra paperwork.
7. Prove & Harden
A joint Pinsaar-bank war-room executed three load campaigns: • 500 TPS sustained for 6 hrs • 800 TPS burst for 5 min • One-DC blackout with live traffic Pen-testers attempted SQLi, privilege escalation, and API fuzzing; no critical vulnerabilities surfaced. Mean failover time: < 200 ms, invisible to frontend clients.
Passing the gauntlet unlocked final sign-off from Risk and CISO, and qualified the platform for inclusion in the bank's Gold Disaster-Recovery tier—normally reserved for core banking and payment switches.
8. Go-Live & Impact
The system flipped on a Friday before the Diwali mega-sale. Traffic hit 20× the old peak, yet p95 latency stayed under 180 ms. Branch dashboards showed "green" all day; staff nicknamed it the "quiet counter" release. Marketing used the newfound capacity to launch instant-approval offers the following week, adding ₹140 crore to the loan book in 30 days.
Customers experienced near-instant approvals; abandonment on the loan application page dropped from 27 % to 4 %. Internal CSAT surveys jumped 22 points as staff moved from firefighting to advisory selling.
9. Lasting Value
With click-to-deploy pipelines, new loan products (e.g., green energy financing) now roll out in 4–7 days instead of 3–4 months. The bank has recorded zero planned downtime and just 18 minutes of unplanned outage in the first year—both times due to upstream provider glitches automatically remediated. Operating cost per transaction fell by 37 %.
The platform became the blueprint for two additional modernisation programs inside the bank and is cited in annual reports as a key driver of digital growth. For Pinsaar, it's living proof that automation-led engineering can compress enterprise timelines without compromising governance.